MCLACHLAN2021 - Bayesian dynamic sound localization model

Usage

[results,template,target] = mclachlan2014(template,target,'num_exp',20,'sig_S',4.2);

Input parameters

template.fs: sampling rate (Hz)
template.fc: ERB frequency channels (Hz)
template.itd0: itd computed for each hrir (samples)
template.H: Matrix containing absolute values of HRTFS for all grid points
template.coords: Matrix containing cartesian coordinates of all grid points, normed to radius 1m
template.T: angular template for each coordinate
target.fs: sampling rate
target.fc: ERB frequency channels
target.itd0: itd corresponding to source position
target.S: sound source spectrum
target.H: Matrix containing absolute values of HRTFS for all source directions
target.coords: Matrix containing cartesian coordinates of all source positions to be estimated, normed to radius 1m
target.T: angular template for each coordinate

Output parameters

`doa`	directions of arrival in spherical coordinates, contains the fields '.est' (estimated DOA [num_sources, num_repetitions, 3]) and '.real' (actual DOA [num_sources, 3])
`params`	additional model's data computerd for estimations

Description

'params' contains the following fields:

'.est_idx Indices corresponding to template direction where': the maximum probability density for each source position is found

'.est_loglik Log-likelihood of each estimated direction'

'.post_prob Maximum posterior probability density for each target source'

'.freq_channels Number of auditory channels'

'.T_template Struct with template data elaborated by the model'

'.T_target Struct with target data elaborated by the model'

'.Tidx Helper with indexes to parse': the features from T and X

MCLACHLAN2021 accepts the following optional parameters:

`'num_exp',num_exp`	Set the number of localization trials. Default is num_exp = 500.
`'SNR',SNR`	Set the signal to noise ratio corresponding to different sound source intensities. Default value is SNR = 75 [dB]
`'dt',dt`	Time between each acoustic measurement in seconds. Default value is dt = 0.005.
`'sig_itd0',sig`	Set standard deviation for the noise on the initial itd. Default value is sig_itd0 = 0.569.
`'sig_itdi',sig`	Set standard deviation for the noise on the itd change per time step. Default value is sig_itdi = 1.
`'sig_I',sig`	Set standard deviation for the internal noise. Default value is sig_I = 3.5.
`'sig_S',sig`	Set standard deviation for the variation on the source spectrum. Default value is sig_S = 3.5.
`'rot_type',type`	Set rotation type. Options are 'yaw', 'pitch' and 'roll'. Default value is 'yaw'.
`'rot_size',size`	Set rotation amount in degrees. Default value is rot_size = 0.
`'stim_dur',dur`	Set stimulus duration in seconds. Default value is stim_dur = 0.1.

Further, cache flags (see amt_cache) can be specified.

Description:

mclachlan2021(...) is a dynamic ideal-observer model of human sound localization, by which we mean a model that performs optimal information processing within a Bayesian context. The model considers all available spatial information contained within the acoustic signals encoded by each ear over a specified hear rotation. Parameters for the optimal Bayesian model are determined based on psychoacoustic discrimination experiments on interaural time difference and sound intensity.

Requirements:

SOFA API v1.1 or higher from http://sourceforge.net/projects/sofacoustics for Matlab (e.g. in thirdparty/SOFA)

References:

R. Barumerli, P. Majdak, R. Baumgartner, J. Reijniers, M. Geronazzo, and F. Avanzini. Predicting directional sound-localization of human listeners in both horizontal and vertical dimensions. In Audio Engineering Society Convention 148. Audio Engineering Society, 2020.

R. Barumerli, P. Majdak, R. Baumgartner, M. Geronazzo, and F. Avanzini. Evaluation of a human sound localization model based on bayesian inference. In Forum Acusticum, 2020.

J. Reijniers, D. Vanderleist, C. Jin, C. S., and H. Peremans. An ideal-observer model of human sound localization. Biological Cybernetics, 108:169--181, 2014.

G. McLachlan, P. Majdak, J. Reijniers, and H. Peremans. Towards modelling active dynamic sound localisation based on Bayesian inference. Acta Acustica, 2021.

THE AUDITORY MODELING TOOLBOX